Detecting surface defects on PCBs is vital for quality inspection. To tackle the problems of low accuracy and poor real-time performance in PCB defect detection, a YOLOv8-based method is proposed. In the feature extraction stage, LP-C2f is employed instead of C2f to improve information extraction efficiency, reduce network redundancy, and accelerate speed. Additionally, by integrating the MPCA attention mechanism and feature fusion module, multi-dimensional attention is focused on relevant feature information to effectively distinguish defects from the background. The introduction of CFPN enhances multi-level information interaction and fusion, focusing on small target detection. Finally, a balanced performance score ( \(BPS\) ) comprehensive evaluation metric is proposed to assess the model’s speed and accuracy. Experimental results demonstrate that the model achieves an \(mAP\) of 98.7%, a 1% improvement over the original YOLOv8s model, with a 60% reduction in model parameters. The FPS value reaches 86, and the comprehensive indicator \(BPS\) is superior, meeting the needs of industrial production.